Deep Learned Process Parameterizations Provide Better Representations of Turbulent Heat Fluxes in Hydrologic Models

Abstract

The debates amongst the hydrologic modeling community about the use and utility of machine learning (ML) to simulate hydrologic processes indicate that much work remains to be done to understand the role and potential of ML in hydrologic modeling (Nearing et al., 2020;Shen, 2018). While it is true that deep learning (DL) models have shown great promise and superior performance in many cases it is yet unclear how to make models that are both composable (i.e., easy to combine with other models) and transferable for scientific studies (i.e., the same model configuration can be used to explore disparate scientific questions). In this paper, we outline an approach for coupling DL models of individual processes into existing hydrologic modeling frameworks. This coupling approach allows us to represent individual physical processes within a larger model using ML methods and to introduce feedbacks between model components. The ability to couple model components will address these composability and transferability questions, as well as allow use of these types of machine-learned models in areas which do not have readily available training data.

Exploring the Potential of Long Short‐Term Memory Networks for Improving Understanding of Continental‐ and Regional‐Scale Snowpack Dynamics

“…In fact, this can be advanced by symbiotic integration of physically based and data-based models that facilitate the development of physics-AI synergy modeling approaches (Reichstein et al, 2019). Recent attempts have included replacing internal process equations with networks that have the ability to learn from data (Bennett & Nijssen, 2021a), the embedding of physically based representations into ML networks (Jiang et al, 2020), and the imposition of mass balance constraints into ML (Hoedt et al, 2021;Nearing et al, 2021).…”

Source Relationships and Model Structures Determine Information Flow Paths in Ecohydrologic Models

“…We use Ta , RH , LE , and H observations from 60 selected FLUXNET2015 sites for which SUMMA is locally calibrated and simulations driven by the flux tower weather data are available (Bennett & Nijssen, 2021b). We then calculate B from daily average Ta and RH using the SFE and from daily average LE and H using SUMMA outputs for each of the three variants to compare model performance against B derived from observations at each site.…”

“…For the land‐surface model, we derive B using existing simulations of H and LE from the Structure for Unifying Multiple Modeling Alternatives (SUMMA, Clark et al., 2015), a modular process‐based hydrologic modeling framework, here, set up to mimic the Noah land surface model (see details in Bennett & Nijssen, 2021a). SUMMA estimates B by solving a complex series of energy and water balance equations, requiring numerous model inputs beyond Ta and RH , including land‐surface parameters and additional weather forcing variables.…”

Can machine learning accelerate process understanding and decision‐relevant predictions of river water quality?

“…Assimilation of new data into models using methods such as ensemble Kalman filters and autoregression (Brajard et al, 2020; Nearing, Klotz, et al, 2021; Zwart et al, 2021), and the use of integrated datasets tailored for the problem can improve prediction outcomes. Software that synthesize data for on‐demand queries such as brokering‐based tools (Horsburgh et al, 2016; Varadharajan et al, 2022), and methods to streamline quality control and outlier detection, gap‐fill, downscale observations, and determine parameters for process models (Bennett & Nijssen, 2021; Campbell et al, 2013; Hill & Minsker, 2010; Leigh et al, 2019; Mital et al, 2020; Russo et al, 2020) would ideally be integrated into ML workflows in parallel with advances in modelling approaches.…”